Why Does Google Dominate Search? The Evidence

By Bill Hartzer · Updated 17 September 2026

The short answer

Google dominates search because two separate things are true at once, and each is supported by a different class of evidence. Contracts that make Google the default — the engine a browser or phone queries when the user types into the address bar without choosing anything — accounted for roughly 50 percent of all United States general search queries in 2020. And Google sees a range of distinct questions that no rival comes near to seeing, which is an advantage in result quality that no contract manufactured.

Neither mechanism explains the position on its own, and the interesting part is that they are not independent. Distribution supplies queries; queries supply the data that improves results; better results make the default feel like the right one, and make the next distribution contract worth paying for. That loop is what the evidence below describes from four directions.

The scale being explained is worth fixing first. StatCounter Global Stats put Google at 91.1 percent of worldwide search referrals in August 2026. The UK Competition and Markets Authority, in its 2025 final decision on the strategic market status investigation into Google's general search services, found on data compelled from Google that Google Search accounts for more than 90 percent of all general search queries in the United Kingdom. The first of those is a panel estimate of referral traffic and the second is a legally binding determination about queries; they are different measurements that happen to land in the same region.

What follows uses sources that can compel disclosure or that randomize people into treatment groups, because almost everything written about this question uses neither. The Department of Justice's trial demonstratives and the district court's memorandum opinion in United States of America v. Google LLC, No. 1:20-cv-03010-APM, rest on Google's own internal query data produced in discovery. The National Bureau of Economic Research working paper cited here paid people to leave Google and watched what they did. Those are the only two ways to get past the circularity that ruins the ordinary version of this argument, in which Google is popular because it is good and is known to be good because it is popular.

One caveat applies to every court figure on this page and is not repeated each time it appears: the liability ruling against Google is on appeal, and the query shares in the record describe 2020, not today.

Roughly half of US queries are placed by contract

The most useful single number about Google's position is not its market share. It is the share of queries that arrive through a paid placement rather than through a decision.

The Department of Justice Antitrust Division's closing argument presentation in U.S. and Plaintiff States v. Google LLC, a trial demonstrative filed in 2024, decomposes it. Google's default search placement contracts covered roughly 50 percent of all United States general search queries in 2020. The deck then breaks that half into its three distribution channels:

The Apple Internet Services Agreement — 28 percent of all US general search queries. One contract with one company accounts for more than half of the contracted total.
Android manufacturer and carrier agreements — 19.4 percent. These are the arrangements with handset makers and mobile network operators that determine what a new Android phone searches with out of the box.
Third-party browser agreements — 2.3 percent. The smallest channel by query volume, and, as the Mozilla accounts discussed elsewhere on this site show, the one where the money matters most to the recipient.

That decomposition is the part almost never reported. Press coverage of the case quotes the 50 percent aggregate and stops, which leaves the reader unable to see that the Apple agreement is not one deal among many but the single largest distribution fact in Western search. The per-channel splits come from Google's internal query data produced in discovery, which is why they exist at this granularity at all — no panel, tracker or survey can measure which queries arrived via which contract.

Two limits belong with the figures. They describe 2020 query data presented at trial in 2024, so they are a snapshot of a market that has since absorbed a generative-answer product on every major surface. And the underlying liability ruling is on appeal, which affects the legal conclusions drawn from the evidence rather than the evidence itself: the query shares were compiled from Google's own records and were not seriously contested.

What the number does not say is also worth stating. Half of US queries arriving by contract does not mean that half of US searchers would leave if the contracts ended. A default is a starting position, not a cage; changing it takes a few taps, and this site documents where the setting lives in every major browser. The question of how many people would actually move is exactly what the next two sections answer, and the answer is not intuitive in either direction.

The Microsoft Edge natural experiment

The strongest single piece of evidence that defaults rather than quality decide search share is a comparison that nobody designed. It sits in the district court's 2024 memorandum opinion in United States of America v. Google LLC, and it takes one sentence to state.

On Microsoft Edge — Microsoft's own browser, where Microsoft sets the default — Bing's search share is approximately 80 percent and Google's is approximately 20 percent. Everywhere else, the ratio is the mirror image of that. The same court found Bing at 5.5 percent of all US general search queries in 2020 against Google at nearly 90 percent.

A natural experiment is a situation in which something outside the researcher's control assigns people to different conditions, allowing a comparison that a controlled trial would otherwise be needed to produce. Edge is one. The people using Edge are not a different species from the people using Chrome or Safari. They did not select the browser because they preferred Bing's results; most of them did not select the browser at all, since Edge ships with Windows. The only systematic difference between the Edge population and everyone else is which engine the address bar queries when nobody intervenes. Flip that one variable and an engine at roughly a twentieth of national query share becomes the engine handling four searches in five.

The magnitude is what makes the observation hard to argue with. If Google's share reflected a large and widely perceived quality gap, Edge users would be expected to override the default in bulk, because overriding it is trivial and Edge has for years placed the search-engine setting a few clicks from the address bar. Instead, the default wins by a factor of four on Microsoft's own surface, in the same country, in the same year, among the same general population.

The honest reading is narrower than the headline. This does not prove that Bing and Google are equal in quality; it proves that whatever quality difference exists is not large enough to overcome a default for most people most of the time. Those are different claims, and the evidence in the last two sections of this page bears on the first one directly. It is also a US finding from a 2024 opinion describing earlier data, on a judgment under appeal.

What the Edge figure does establish is the price mechanism behind the contracts in the previous section. If defaults did not move share, nobody would pay for them. They move share by roughly this much, which is why the payments discussed on the default-placement page reach the scale they do.

What happened when researchers paid people to leave Google

Observational evidence cannot separate preference from habit, because the people who use Google have never been made to try anything else. One study solves that by paying them to.

Sources of Market Power in Web Search: Evidence from a Field Experiment, National Bureau of Economic Research working paper 33410, is a randomized field experiment in which participants were offered money to switch their search engine to Bing for two weeks. It should be named for what it is: a working paper, first issued in January 2025 and revised in February 2026, not a peer-reviewed article. Confirm its publication status before treating its estimates as settled.

The baseline is stark. 96 percent of participants used Google for the majority of their searches. Against that, three results matter.

Money moves people. Offered $10 to use Bing for two weeks, 58 percent switched. The paper's price ladder records Bing usage rising with the size of the payment, reaching roughly 31 percent at $1, 64 percent at $10 and 74 percent at $25. The $10 figures differ between the two measures because acceptance of the offer and sustained usage are not the same quantity.
A third of them stayed. 33 percent were still using Bing after the incentive ended. Nobody was paying them at that point. They had tried the alternative, and enough of them preferred it — or found the difference small enough not to bother switching back — that the share persisted.
Beliefs moved with experience. Before trying Bing, 61 percent of users said Google was a lot better. After two weeks of forced use, perceptions shifted by 0.6 standard deviations, a substantial movement in a measure of stated preference.

That 33 percent is the most informative number on this page, and it cuts against both of the comfortable stories. It is far too high to be consistent with the claim that Google's dominance simply reflects a product so much better that no rational person would use anything else; a third of a randomly selected group, given two weeks of enforced exposure, did not go back. And it is far too low to be consistent with the claim that share is purely an artifact of placement, because two-thirds returned to Google the moment they were free to.

The sample limits the reach of the finding. Participants were recruited from the Prolific research panel and were desktop users, a population younger and more technically capable than the general public, and desktop is the device class where Google's position is weakest. A mobile-first sample would plausibly show more inertia rather than less.

Choice screens barely move anything

If defaults place half the queries, the obvious remedy is to stop defaulting: show the user a screen listing several engines and let them pick. Regulators have reached for this repeatedly, and it underpins both the European Union's Digital Markets Act approach and parts of the United States remedies debate. The same NBER working paper measured what a choice screen — an interstitial that presents a list of search engines and requires an explicit selection — actually does.

The screens in question typically list a handful of engines — Bing, DuckDuckGo and Ecosia are the names most often presented alongside Google — and ask the user to pick one before proceeding. In a two-month field experiment with 2,354 desktop users, only 1.2 percent of Google users shown an active choice screen switched to Bing. Modeling out from that result, the authors estimate that choice screens would increase Bing's market share by approximately 1.3 percentage points.

That is close to nothing, and the reason it is close to nothing is visible in the same paper's other findings. People who have never used an alternative believe, by a large majority, that Google is substantially better; 61 percent said so before trying Bing. A choice screen asks a user to act on a belief, and the belief is wrong in a way that only experience corrects. Two weeks of forced use moved perceptions by 0.6 standard deviations; a list of names on a screen moves them by nothing at all, because the user learns nothing from reading it.

This finding sits awkwardly with the remedies most often proposed, and the awkwardness is worth stating rather than smoothing over. The evidence suggests that the intervention regulators favor least — compelled trial, in which a user is made to run an alternative for a period — is the one that changes behavior, while the intervention they favor most, an informed choice at the point of setup, changes almost none of it. Whether compelled trial is a proportionate remedy is a question this page does not attempt to answer; it is a legal and political judgment, not an empirical one.

The same caveats apply as before. This is a working paper. The panel is Prolific desktop users, skewing young and technical, which if anything should make the estimated switching rate an upper bound rather than a floor: a sample more comfortable with changing software settings than the general population still moved by 1.2 percent. And the experiment tested a choice screen presented to existing Google users, not one presented at first device setup to someone with no incumbent habit, where the effect could differ.

The part that is not habit: query diversity and data scale

Everything above concerns how queries get to Google. This section concerns what happens once they arrive, and it is the part of Google's advantage that no contract created and no remedy easily reverses.

Query diversity — the range of distinct questions an engine has ever been asked, as distinct from the volume of questions it handles — is the quantity that compounds. The trial record measured it. Across a seven-day sample of 3.7 million unique search phrases, 93 percent were seen only by Google, while just 4.8 percent were seen only by Bing.

Read that against share and the asymmetry is larger than it first appears. An engine holding roughly a twentieth of query volume does not encounter a twentieth of the questions. It encounters most of the common ones and almost none of the rare ones. Rare queries are precisely where an engine has the least evidence about which result is good, because there are no prior sessions to learn from, and they are where the quality difference between a large index with behavioral signals and a smaller one is most visible to a user. The gap does not close by winning users at the margin, because a marginal user brings mostly common queries that the smaller engine already sees.

The Department of Justice's closing demonstrative puts the same point in aggregate: Bing has approximately 5 percent of Google's user-side search data. That estimate is derived from the trial record, Google disputes the framing, and the case is on appeal — but even taken loosely, a twentyfold data gap is not something a competitor closes by shipping a better interface.

This is the strongest available answer to the question of whether Google's results are actually better, and it is an indirect one. Nobody has run a large blind quality comparison between the major engines that would settle it directly. What the record supports is the mechanism: Google observes vastly more of the question space and vastly more behavior against it, and an engine that observes more has more to rank on. The NBER experiment's 33 percent retention figure suggests the resulting difference is smaller than users believe it to be, which is a separate finding and does not contradict this one.

The query-diversity figure is reproduced in a University of Chicago Law School course supplement, Picker, Antitrust Fall 2025, Supplement 4, which reprints the court's opinion. Where a citation of record is required, cite the primary docket filing rather than the supplement.

Neither purely earned nor purely bought

The evidence does not support either of the two positions people usually hold.

It does not support the view that Google's share is simply the market rewarding a better product. Half of US queries in 2020 arrived through paid placement rather than choice. Flipping the default on one browser inverts the outcome, giving Bing 80 percent where it otherwise holds 5. And more than half of a randomized sample walked away from Google for $10, which is not the behavior of people who believe they are using something irreplaceable.

It does not support the view that the position is purely bought, either. A third of the people who were paid to try Bing were still using it when the money stopped, but two-thirds returned. Choice screens move share by roughly a percentage point. And the data-scale advantage is real, large and independent of any contract: 93 percent of a seven-day sample of unique phrases were seen by Google and nobody else, which is a property of the system that would persist for years even if every distribution agreement were voided tomorrow.

What the evidence actually describes is a loop. Contracts place queries. Queries produce behavioral data at a scale no competitor can match. That data improves results, particularly on the rare and ambiguous queries where the difference is perceptible. Better results make users content with the default and make advertisers pay more per query, which funds the next round of placement payments, which places more queries. Each mechanism makes the other cheaper to sustain.

The consequence is that asking whether Google's dominance is deserved is not a well-formed question. The distribution and the quality are not separable inputs that could be weighed against each other, because each has been producing the other for the better part of two decades. A useful version of the question is narrower and answerable: what would happen to result quality if the distribution stopped? The record contains no answer to that, because nobody has ever observed it.

Readers who want the commercial half of this loop in detail will find the payment figures on the default-placement page: the court found that Google paid more than $26 billion in revenue-share payments for default placement in 2021, nearly four times more than all of its other search-specific costs combined.

What this evidence does not settle

Four limits are worth carrying away with the numbers.

The court figures are 2020 vintage and under appeal. Both the 50 percent contracted-query share and the Edge comparison describe a market before generative answers appeared on every major search surface. The liability ruling has been appealed and no appellate decision had issued as of this page's last check.
The field experiment is a working paper. NBER working paper 33410 has not completed peer review, and its sample is a panel of desktop users recruited through Prolific — younger and more technically confident than the population, on the device class where Google is weakest.
No direct quality comparison exists. Nothing in the record establishes how much better Google's results are than Bing's, or whether they are better at all on ordinary queries. The data-scale evidence establishes an input advantage, not an output one, and the 33 percent retention figure is the closest thing to a revealed-preference measurement anybody has produced.
The AI era is unmeasured on these terms. Every figure here concerns conventional search queries. What happens to the distribution loop when a meaningful share of questions are answered by an assistant with no default relationship to a browser is not something any of these sources measured, and estimates of it are currently modeling rather than observation.

None of that undermines the core finding, which is unusually well evidenced by the standards of writing on this subject: Google's position rests on distribution it pays for and on a data advantage that distribution built, and both are large.

Frequently asked questions

Why is Google the most used search engine?

Two mechanisms, both evidenced. Distribution: the Department of Justice's trial demonstrative in U.S. and Plaintiff States v. Google LLC records that Google's default placement contracts covered roughly 50 percent of all US general search queries in 2020, with the Apple Internet Services Agreement alone covering 28 percent. Scale: the trial record shows that across a seven-day sample of 3.7 million unique search phrases, 93 percent were seen only by Google, an advantage in data that no contract created. The two reinforce each other, which is why they are hard to separate.

How much of Google's search traffic comes from being the default?

Roughly half of US general search queries in 2020, per the Department of Justice's closing argument demonstrative, broken down as 28 percent through the Apple Internet Services Agreement, 19.4 percent through Android manufacturer and carrier agreements, and 2.3 percent through third-party browser agreements. That is the share arriving through a paid placement, not the share that would leave if the placements ended. Those are different quantities and the second one has never been observed.

Would people switch away from Google if they had a real choice?

Some would, and fewer than the argument usually assumes. In a randomized field experiment reported in National Bureau of Economic Research working paper 33410, 96 percent of participants used Google for most of their searches; offered $10 to use Bing for two weeks, 58 percent switched, and 33 percent were still using Bing after the payments stopped. So about a third of people made to try the alternative preferred it enough to stay, and two-thirds went back. It is a working paper on a panel of desktop users, so treat the figures as indicative rather than settled.

Do choice screens reduce Google's market share?

Barely. The same NBER working paper ran a two-month field experiment with 2,354 desktop users and found that only 1.2 percent of Google users shown an active choice screen switched to Bing, with the authors' model predicting that choice screens would raise Bing's share by approximately 1.3 percentage points. The likely reason is that a choice screen asks users to act on a belief about relative quality, and 61 percent of them believed Google was a lot better before ever trying an alternative. Two weeks of forced use shifted those perceptions by 0.6 standard deviations; reading a list of names shifts them by nothing.

Is Google actually better than Bing, or do people just use it out of habit?

No source settles this, because no large blind comparison of result quality has been published. What the record supports is that Google's inputs are far larger: 93 percent of a seven-day sample of 3.7 million unique phrases were seen only by Google, and the Department of Justice's demonstrative estimates Bing at approximately 5 percent of Google's user-side search data. What the randomized evidence suggests is that the perceived gap is bigger than the experienced one, since a third of people paid to try Bing kept using it unpaid. Both things can be true at once.

What does Microsoft Edge prove about search defaults?

It is the clearest natural experiment available. The district court's 2024 memorandum opinion records that Bing holds approximately 80 percent search share on Microsoft Edge against Google's 20 percent, the mirror image of the ratio everywhere else, while the same court found Bing at 5.5 percent of all US general search queries in 2020. The Edge population is not unusual; the only systematic difference is who sets the default. It shows that whatever quality gap exists is not large enough to make most people override a default, which is a narrower claim than saying the engines are equally good.

Could another search engine catch up with Google?

Not by winning users at the margin, on the evidence in the trial record. The obstacle is query diversity rather than query volume: 93 percent of a seven-day sample of unique search phrases were seen only by Google, and a competitor gaining share picks up mostly common queries it already sees, not the rare ones where it has no evidence to rank on. The Department of Justice's estimate that Bing holds approximately 5 percent of Google's user-side data describes the size of that gap. Google disputes the framing and the case is on appeal.

Are these figures current?

The court and Department of Justice figures describe 2020 query data presented at trial in 2024, and the liability ruling is on appeal. The NBER field experiment was first issued in January 2025 and revised in February 2026 and remains a working paper. For current share, StatCounter Global Stats put Google at 91.1 percent of worldwide search referrals in August 2026, and the UK Competition and Markets Authority found in 2025 that Google accounts for more than 90 percent of UK general search queries. None of the sources on this page measures what generative answer products have done to the distribution loop they describe.

Why Google dominates search, on the evidence